sharks.Rmd

---
title: "Calibrating a molecular phylogeny of the sharks"
date: "`r format(Sys.time(), '%b %d, %Y')`"
bibliography: bibliography.bib
output:
  html_document:
    highlight: tango
    code_folding: show
    theme: sandstone
    toc: true
    toc_depth: 3
    toc_float:
      collapsed: false
---

# Background
The time-calibration of molecular phylogenies is essential to many phylogenetic comparative analyses.
Time-calibration can be accomplished with e.g. data from fossil specimen that can be assigned to a certain 
taxonomic group. Specimen ages can be determined by
e.g. carbon dating or stratigraphic methods. Node ages are usually assigned using Bayesian or
maximum-likelihood methods.

Some of the specimen records at Naturalis hold stratigraphic information.
Here we will demonstrate how to extract this data using `nbaR` and
how to create input for the popular tree-calibration function `chronos` from the phylogenetic analysis
package `ape`.

## The phylogeny
As input, we will use a species-level molecular shark phylogeny published by
[@VELEZZUAZO2011207].

The phylogeny is a majority-rule consensus tree inferred from molecular markers using
Bayesian inference and comprises 229 species in all eight orders of the sharks (superorder *Selachimorpha*):

![](https://ars.els-cdn.com/content/image/1-s2.0-S1055790310004537-gr3.jpg)

The non-ultrametric tree in the above figure is not time-calibrated, the branch lengths thus
represent molecular distances.

# Getting chronological data with nbaR
Chronological association is associated with `Specimen` data, we thus instantiate
a `SpecimenClient`:

```{r load_nbaR}
library(nbaR)
sc <- SpecimenClient$new()
```

In total, there are [eight extant shark orders](https://en.wikipedia.org/wiki/Shark).

```{r}
shark_orders <- c(
  "Carcharhiniformes",
  "Heterodontiformes",
  "Hexanchiformes",
  "Lamniformes",
  "Orectolobiformes",
  "Pristiophoriformes",
  "Squaliformes",
  "Squatiniformes"
)
```

We can then formulate a query condition for specimens that are identified in one of these orders
using the operator `IN`:

```{r qc_1}
qc <-
  QueryCondition$new(field = "identifications.defaultClassification.order",
                     operator = "IN",
                     value = shark_orders)

```

For many specimens, chronological information is available in the fields `gatheringEvent.chronoStratigraphy.youngChronoName`
and `gatheringEvent.chronoStratigraphy.oldChronoName` which represent upper- and lower time bounds, respectively.
Below, we will formulate a `QueryCondition` that requires either one of these fields to be non-empty:

```{r qc_2}
## formulate query conditions for fields to be non-empty
qc2 <-
  QueryCondition$new(field =
                       "gatheringEvent.chronoStratigraphy.youngChronoName",
                     operator = "NOT_EQUALS",
                     value = "")
qc3 <-
  QueryCondition$new(field =
                       "gatheringEvent.chronoStratigraphy.oldChronoName",
                     operator = "NOT_EQUALS",
                     value = "")

## join qc2 and qc3 with operator OR
qc2$or <- list(qc3)
```

Now we can do the query:

```{r do_query}
## instantiate QuerySpec, give size
qs <- QuerySpec$new(conditions = list(qc, qc3), size = 5000)
res <- sc$query(querySpec = qs)

## how many hits?
res$content$totalSize
```

**Note**: By default, a query returns only data for the first 10 hits. 
Above, we set the `size` parameter to the QuerySpec constructor to allow the download of 
up to 5000 values This number was based on the prior knowloedge that there are less
than 5000 records for our query; We advise to always check the length of the resultSet
against the `totalSize` of a `QueryResult` to make sure that all records are downloaded, 
e.g.

```{r check_length}
length(res$content$resultSet) == res$content$totalSize
```

## Exploring the data

Now we can explore the data:

```{r explore}
## load all specimens
specimens <- lapply(res$content$resultSet, function(x)
  x$item)

## check the fields youngCronoName and oldChronoName
unique(unlist(
  lapply(specimens, function(x)
    x$gatheringEvent$chronoStratigraphy[[1]]$youngChronoName)
))
unique(unlist(
  lapply(specimens, function(x)
    x$gatheringEvent$chronoStratigraphy[[1]]$oldChronoName)
))
```

**Caution**: As you see above, there can be more than one chronoStratigraphy assigned with a
specimen. Check how many we have per specimen:

```{r num_chrono}
unique(sapply(res$content$resultSet, function(x)
  length(x$item$gatheringEvent$chronoStratigraphy)))
```

Each `gatheringEvent` in each `Specimen` has only one `chronoStratigraphy`. We thus do not miss data by taking
the first one (`chronoStratigraphy[[1]]`).

Do we have hits for all orders? Below, we make an overview of the field `identifications.defaultClassification.order`
for all specimen.

```{r plot_orders}
## make table with counts for each order
tab <-
  table(unlist(
    lapply(specimens, function(x)
      x$identifications[[1]]$defaultClassification$order)
  ))
par(mar = c(5.1, 8, 4.1, 2.1))
barplot(sort(tab),
        horiz = TRUE,
        las = 2,
        xlab = "Number of specimens")
```

**Caution**: There can be multiple identifications for one specimen. Some specimens 
in our data are for instance assigned to different genera. However, 
the field `preferred` in an `Identification` object states if this is the most accurate and 
recent identification. The preferred identification is usually the first in the list of a specimen's identifications.
It can, however, occur that there are multiple identifications of which none is preferred; it is best to filter them out:

```{r preferred}
## get the specimens which have a preferred identification
ident <-
  sapply(specimens, function(s)
    any(sapply(s$identifications, function(i)
      i$preferred)))

## how many do not have a preferred identification?
sum(!ident)

## filter specimens
specimens <- specimens[ident]
```

Let's further explore the data. To see what part of the animals were preserved, we can look into
the field `kindOfUnit`:

```{r kindofunit}
table(sapply(specimens, `[[`, "kindOfUnit"))
```

Almost all of them are teeth.

## Getting absolute ages
The type of chronostratigraphic data (as shown above) is given in divisions on the geological time scale, e.g. *Miocene*.
These strings can refer to *eons*, *eras*, *periods*, *epochs* and *ages*. In order to
translate this into absolute ages, we will use the API of the [Earth Life Consortium](http://earthlifeconsortium.org/).
This is possible using the *convenience function* `geo_age`. For example

```{r geo_age}
## get the upper chrono name for the first specimen
name <- specimens[[1]]$gatheringEvent$chronoStratigraphy[[1]]$oldChronoName
name

## get the lower and upper bounds for this division
geo_age(name)
```

We can now make a table with all interesting data, below this is done for the
first 50 specimen objects:

```{r chrono_name}
## Get genus, species and chrono information from specimen records
data <-
  as.data.frame(do.call(rbind, lapply(specimens[1:50], function(x) {
    genus <- x$identifications[[1]]$defaultClassification$genus
    if (is.null(genus))
      genus <- NA
    specificEpithet <-
      x$identifications[[1]]$defaultClassification$specificEpithet
    if (is.null(specificEpithet))
      specificEpithet <- NA
    youngChronoName <-
      x$gatheringEvent$chronoStratigraphy[[1]]$youngChronoName
    if (is.null(youngChronoName))
      youngChronoName <- NA
    oldChronoName <-
      x$gatheringEvent$chronoStratigraphy[[1]]$oldChronoName
    if (is.null(oldChronoName))
      oldChronoName <- NA
    c(
      genus = genus,
      specificEpithet = specificEpithet,
      youngChronoName = youngChronoName,
      oldChronoName = oldChronoName
    )
  })))

## Get absolute ages from earth life consortium
times <-
  geo_age(unique(c(
    as.character(data$youngChronoName),
    as.character(data$oldChronoName)
  )))

## add upper and lower bounds to ages
data$young_age <-
  sapply(data$youngChronoName, function(x)
    ifelse(is.na(x), NA, unlist(times["late_age", as.character(x)])))
data$old_age <-
  sapply(data$oldChronoName, function(x)
    ifelse(is.na(x), NA, unlist(times["early_age", as.character(x)])))

data
```

# Calibrating the phylogeny

Depending on the data, calibration points could be chosen on different taxonomic levels.
If there are sufficient specimens determined at species level, one could use the upper- and
lower bounds above. It is also possible to average upper- and lower values for a higher taxonomic
group, such as genus or family. Since this requires some data cleaning, such as dealing with duplicates,
missing data, etc,
`nbaR` offers the function `chronos_calib` which takes a set of specimen object, and a tree and
averages the data for a user-defined taxonomic
group and returns a calibration table that can be directly used as input for the function `chronos`
from the package `ape`. The function also determines the node which will be calibrated in the phylogenetic tree.

The shark phylogeny comes with the package and can be
parsed using the package `ape`:

```{r read_tree}
library(ape)

## read data
data(shark_tree)

## plot the tree
plot(shark_tree, cex = 0.3)
```

## Genus level

Now we use `chronos_calib` to get the table at the genus level. `chronos_calib`
also selects the nodes in the tree that will be calibrated. This is done by
selecting the most recent common ancestor (mrca) of the species for which the data are averaged.

```{r chronos_calib, warning=FALSE}
## make calibration table on genus level
calibration_table <- chronos_calib(specimens, shark_tree, "genus")
```

**Note**: This function can produce many warnings, a warning is emitted
whenever, for some geological division, no data can be obtained from the
earth life consortium. This can be due to misspellings or words in foreign languages.

The calibration table looks as follows:

```{r table}
calibration_table
```


**Note**: It is essential to thoroughly evaluate the calibration table! In this example,
for example, the genus *Squatina* is non-monophyletic and one species is
placed somewhere else in the tree. Calibrating the most recent common ancestor node
for this genus can therefore result in errors using the calibration
routine. We will therefore remove the calibration points for genus *Squatina* before calibration:


```{r plot_calib}
## clean up: one rogue taxon in genus "Squatina"! Skip this genus
calibration_table <-
  calibration_table[calibration_table$taxon != "Squatina",]

## run ape's chronos
chronogram <- chronos(shark_tree, calibration = calibration_table)

## plot tree with time axis
plot(chronogram, cex = 0.3)
axisPhylo()

## plot calibrated genera
nodelabels(calibration_table$taxon, calibration_table$node)
```


## Family level

We can also calibrate the tree on the family level:

```{r, calib_family, eval=FALSE}
## make calibration table on family level
calibration_table <- chronos_calib(specimens, shark_tree, "family")

## run ape's chronos
chronogram <- chronos(shark_tree, calibration = calibration_table)

## plot tree with time axis
plot(chronogram, cex = 0.3)
axisPhylo()

## plot calibrated families
nodelabels(calibration_table$taxon, calibration_table$node)
```

# References